Coinbase CEO floats Chinese open-weight AI models as antidote to runaway bills

Coinbase’s CEO has proposed experimenting with cheaper open-weight AI models to keep AI spending in check as token consumption climbs.

This proposal has led to concerns over the security and geopolitical risks of directing enterprise workloads through Chinese-origin systems.

Why are companies using Chinese AI models?

U.S. export controls have made it harder for Chinese companies to access American AI chips, but that hasn’t stopped them from building competitive models and selling them at much lower prices.

For instance, Zhipu’s $GLM 5.2 costs $1.40 per million input tokens and $4.40 per million output tokens compared to Anthropic’s Opus 4.8 at $5 and $25 for the same volume.

$GLM 5.2 scored 62.1 on SWE-bench Pro, a key coding benchmark, beating OpenAI’s GPT-5.5 at 58.6. One AI researcher said $GLM 5.2 “is at least as good as Opus 4.8 and GPT 5.5.”

Another called it “the first open model that can really compete with closed-source systems.”

Is Coinbase using Chinese AI models

Coinbase’s CEO Brian Armstrong says the best way to control rising AI costs is to use cheaper open-weight models, including systems from China like $GLM 5.2.

Armstrong said instead of spending more and more on AI, companies need “better defaults, routing, and caching.” His suggestion to use Chinese models, even if they are cheaper, has drawn concerns about security and political risks.

Beyond its convenient pricing, $GLM 5.2 uses an MIT license, meaning companies can download it, modify it, and run it on their own servers, removing any risk of sending sensitive company data to an outside API.

AI spending has become a genuine issue, causing companies to roll back the use of the technology in operations.

Cryptopolitan recently reported that Uber used up its entire 2026 AI coding budget by April and now caps engineers at $1,500 per tool each month. Meta sent a memo warning of an “exponential increase” in AI usage and started building spending controls. Amazon scrapped an internal leaderboard that ranked employees by AI consumption because people were gaming it and driving costs up.

A KPMG survey found only 26% of companies have full visibility into their AI costs, while 22% discover spending only after receiving the bill. Goldman Sachs projects that AI token consumption could increase 24-fold by 2030, reaching 120 quadrillion tokens per month.

The International Data Corporation predicts 70% of leading AI-driven enterprises will use multiple models by 2028 rather than relying on a single provider.

What makes Chinese AI models risky?

Z.ai’s cloud API, which allows developers and companies to use its AI models (including $GLM 5.2), falls under China’s National Intelligence Law. That raises real concerns for any company handling sensitive information.

U.S. lawmakers opened a formal inquiry in May into cybersecurity risks from Chinese-origin AI models in critical infrastructure.

There are also concerns that models trained under different legal systems could carry undisclosed behaviors. Adding to that, an AI builder tested $GLM 5.2 against GPT-5.5 on a debugging task and found it “not even close” to the OpenAI model’s ability to spot problems, despite reports that Chinese models outperform their more expensive counterparts.

Anthropic disclosed in an open letter to the Senate Banking Committee that Alibaba Qwen operators ran 28.8 million Claude exchanges through about 25,000 fake accounts between April and June. They called it the largest known campaign to steal a model’s capabilities.

Self-hosting the open weights eliminates the API data-routing risk, as companies that run the model on their own servers don’t send data to China. But the concern about the models themselves remains.

Source